Utility of HIV-1 DNA genotype in determining antiretroviral resistance in patients with low or undetectable HIV RNA viral loads

Abstract

Objectives

To investigate the extent to which drug resistance can be evaluated from proviral HIV-1 DNA genotype compared with RNA genotype at different timepoints.

Patients and methods

In HIV-1-infected patients routinely seen at a university hospital, who needed to change their current ART, antiretroviral drug resistance was determined from DNA genotype and was compared with past RNA genotype (group 1) or same-day RNA genotype (group 2). A ‘resistance sum’ was defined as the sum of agents to which resistance was present and was calculated across NRTI, NNRTI and PI. We defined ‘loss of information’ as when a lower resistance sum was observed in DNA than in RNA samples.

Results

Of the 74 and 26 patients included in groups 1 and 2, respectively, most had a long median duration of known HIV-1 infection (17.4 and 14.2 years) and ART (15.3 years and 13.5 years). For group 1, the median (range) resistance sums between DNA/RNA were 0 (0–6)/1 (0–6) for NRTIs, 0 (0–4)/0 (0–4) for NNRTIs and 0 (0–7)/0 (0–8) for PIs, which were comparable with group 2. Loss of information in DNA was substantial for group 1 (37.8%) and less so for group 2 (11.1%). In multivariable analysis, only longer ART duration was significantly associated with loss of information. Results were similar in patients harbouring resistance to one or more agents.

Conclusions

In a real-life setting, genotyping DNA from PBMC has some degree of concordance compared with RNA. Loss of information in DNA would appear to coincide with longer periods of ART.

Introduction

With potent and well-tolerated ART, plasma viral loads (VLs) of HIV can be easily suppressed to undetectable levels for the majority of HIV-infected patients. For example, data from the French Hospital Database on HIV have recently described 90.3% of treated patients from 2009–11 exhibiting virological success.¹ Consequently, the evolution towards severe HIV-related disease has been drastically curtailed in infected individuals undergoing ART.²

Despite these successes, some individuals need to modify their ART regimen owing to virological failure at low levels of VL. Meanwhile, other patients with undetectable HIV plasma RNA might change treatment combinations as a result of toxicity of certain agents or to simplify their ART regimen.³ In order to guide the choice of appropriate treatment, resistance testing would then be recommended. Plasma RNA genotype is the current standard to assess resistance to antiretroviral drugs (ARVs);⁴ however, when VL is low or undetectable, viral sequences from plasma RNA can be difficult to obtain and drug resistance mutation (DRM) results can be unavailable. For instance, only 60% of samples with VL between 20 and 200 copies/mL were successfully sequenced in a nationwide study of on-treatment virological failures conducted in France.⁵ Other ways to determine antiretroviral resistance are needed, specifically in the context of low or undetectable VL.

HIV-1 DNA from samples of PBMCs can be sequenced and could provide some information to assess antiretroviral resistance. However, some previous studies have shown that DNA genotype does not identify all DRMs compared with RNA genotype obtained after ART initiation^6,7 whereas others have found good concordance and that mutations in DNA were predictive of virological failure.^8,9

In this study, we aimed to investigate whether HIV-1 DNA genotype could provide reliable information to guide ARV choice. DRMs and ARV resistance determined from DNA samples were compared with: (i) past RNA genotypes spanning several years; or (ii) same-day RNA genotypes in treated patients changing their current ART regimen. In addition, the net gain or loss of DRM information was examined within classes of ARVs. We also assessed the determinants for which information on antiretroviral resistance is lost when genotyping DNA.

Patients and methods

Study participants

HIV-1-infected individuals seeking care at the Infectious Diseases Department of Saint-Antoine Hospital (Paris, France) were consecutively included between 2010 and 2014 provided that ≥1 DNA genotype and ≥1 RNA genotype were available during clinical follow-up. DNA and RNA genotypes were strictly performed at the request of the treating physician and no selection strategy based on sample availability was used. Two groups were then constructed: patients with undetectable VL or experiencing low-level plasma viraemia, defined as VL <500 copies/mL, whose RNA genotype had been performed prior to DNA genotype (group 1); and patients with VL <1000 copies/mL whose DNA and RNA genotypes were performed on the same day (group 2).

Ethics

Written informed consent was obtained from patients for the use of stored samples and personal data when conducting non-interventional research.

Assessing clinical and virological characteristics

The following patient characteristics were obtained from a computerized medical record database (DIAMM v8.6r0, Villers-lès-Nancy, France): age, sex, mode of transmission, date of HIV-1 diagnosis, AIDS-defining illness and concomitant and nadir CD4+ T cell count. Complete history of ART regimens from treatment initiation until the time of DNA genotyping was collected and verified by the treating physician. VLs were obtained from records at the Virology Department of Saint-Antoine Hospital.

HIV-1 RNA was quantified using Cobas AmpliPrep (Cobas Taqman HIV-1 assay, version 2.0; Roche Diagnostics, France). Total HIV-1 proviral DNA was retrospectively quantified using a real-time PCR assay (LTR HIV-1 DNA assay, Biocentrics, Bandol, France) from the same PBMC samples used to obtain DNA genotypes.¹⁰

Sanger sequencing and assessing DRMs in plasma and PBMC samples

HIV-1 resistance testing was done on routine samples at the Virology Department of Saint-Antoine Hospital. Sanger sequencing was performed on both RNA and DNA for reverse transcriptase (RT) and protease sequences with nested PCRs using primer sequences established by the Agence Nationale de Recherche sur le SIDA (ANRS) consensus method (described at www.hivfrenchresistance.org). RNA was used as template for reverse transcription before amplification. Sequence products were analysed with the GENETIC ANALYZER 3500xl Dx (Applied Biosystems, Villebon Sur Yvette, France) and sequence alignment was performed using Seqscape software (Applied Biosystems). Sequencing of RT is expected to be successful in 60%, 78% and 78% of patients with VLs of 51–200, 201–500 and 501–1000 copies/mL, respectively,⁵ and 297 of 3430 (8.6%) requests for genotyping at our Virology Department during 2010–14 were able to be sequenced. Mutations for NRTI, NNRTI and PI described in the ANRS 2014 v24 algorithm considered to be resistance-associated mutations were recorded (see Table S1 available as Supplementary data at JAC Online).¹¹ Resistance and susceptibility to each agent pertaining to the three classes of ARV were determined by RNA and DNA sequences.

Statistical analysis

All analysis was stratified by study group. Characteristics of each group were summarized using number and percentage for categorical variables and median and IQR for continuous variables.

To assess the similarity of information between samples, we used an indicator of overall resistance defined herein as the ‘resistance sum’. Resistance sum was defined as the number of drugs to which resistance was present and was calculated within classes of ARV (NRTI, NNRTI and PI) and sequence source (RNA or DNA genotype). Only resistance to ARVs was considered in this score, regardless of whether multiple DRMs to the same agent were present. Distributions of resistance sums were truncated to two or more agents in analysis owing to the sparse distribution of sums above this level. DNA and RNA genotypes were compared using percentage agreement and intraclass correlation (ICC). In group 1, data from the most recent RNA genotype was used in analysis for patients with more than one genotype (n = 19). We also conducted a sensitivity analysis among individuals with >1 prior RNA genotype whereby mutation presence was defined whenever it was observed on past RNA genotyping results.

To quantify the amount of information lost or gained with DNA genotype, we calculated a net difference in resistance sum score comparing the number of agents to which resistance was observed between DNA and RNA genotypes. Each agent was assigned the following score: −1 = resistance identified in RNA but not DNA; 0 = resistance identified (or not identified) in both RNA and DNA; and 1 = resistance identified in DNA but not RNA. The sum within individuals was calculated to create a score, which was provided in the overall group and within classes of ARV. Any values <0 and >0 were interpreted, respectively, as loss of and gain in information from DNA genotype compared with RNA. Score distributions were assessed using frequency histograms.

Finally, to understand risk factors associated with the net difference in resistance information, we modelled mean differences in resistance sum score between levels of determinants (Δ) and 95% CI using linear regression. In a separate analysis, we defined a resistance sum score <0, representing loss of information, as an endpoint. ORs and 95% CIs for various demographic, virological and treatment correlates were estimated using logistic regression. For both endpoints, multivariable analysis was performed in which all covariates with P < 0.1 were placed in a full model and removed in backwards-stepwise fashion if the P value was no longer above this threshold.

Statistical analysis was carried out using STATA statistical software (v12.1, College Station, TX, USA) and significance was determined using a P value of <0.05.

Results

Study population

A total of 100 patients met study inclusion criteria for one of the two cohorts: 74 with previous RNA genotypic results prior to DNA genotype (group 1) and 26 with concomitant RNA and DNA genotypes (group 2).

For group 1, 52 patients (70.3%) had undetectable VL (<50 copies/mL). The remaining 22 patients had low level viraemia with a median VL of 1.88 log₁₀ copies/mL. RNA genotypic results were obtained a median 5.8 years prior to DNA genotype in this group, whose distribution was fairly uniform from 0.4 to 10.4 years. Fifteen (20.3%) patients from this group had their RNA genotyping performed before initiating ART. In addition, 19 patients had >1 RNA genotyping prior to DNA genotyping with the number of retrospective RNA genotypes distributed as follows: two (n = 10), three (n = 7), four (n = 1) and five (n = 1). For group 2, the median VL was 2.51 log₁₀ copies/mL for the 24 of 26 patients (92.3%) with detectable VL. Two (7.7%) patients from this group had RNA and DNA genotypes prior to ART initiation.

Patient characteristics for each group are given in Table 1. Patients in groups 1 and 2, respectively, were predominantly male (75.7% and 65.4%), harboured HIV-1 subtype B virus (75.3% and 69.2%) and had a long duration of known HIV-1 infection (median 17.4 and 14.2 years). Patients underwent ART for a median of 15.3 years (group 1) and 13.5 years (group 2), while receiving a median of 9 (IQR 6–12) and 6 (IQR 4–11) ARVs respectively, from ART initiation until the time of DNA genotyping. At the time of DNA genotyping, ART was mostly prescribed in three-drug regimens (n = 53, 52.5%): 2 NRTIs + 1 NNRTI (n = 17), 2 NRTIs + 1 PI (n = 20), 2 NRTIs + 1 integrase strand transfer inhibitor (INSTI) (n = 6) or other combinations (n = 10; 7 of which contained an INSTI). Non-three-drug regimens consisted of monotherapy with ritonavir-boosted darunavir (n = 2), dual therapy (n = 9) or >3 agents from at least three different classes (n = 27; 21 of which contained an INSTI).

Distribution of DRMs in DNA and RNA genotypes

WT genotype in both RNA and DNA was found in 23 patients (31.1%) from group 1 and 9 patients (34.6%) from group 2. Of those with RNA genotyping prior to ART initiation, 11/15 (73.3%) and 1/2 (50.0%) patients from groups 1 and 2, respectively, harboured strains without any antiretroviral resistance.

For patients with DRMs, in group 1, the percentage of DRMs was consistently higher in the most recently performed RNA than DNA genotyping, which was observed for each ARV class (Figure S1A). In contrast, this proportion was similar between sequence sources for all three ARV classes in group 2 (Figure S1B).

Of note, mutations E138K, M184I and M230I in DNA, associated with APOBEC3G/F editing, were found in four patients in group 1 and six patients in group 2. Among them, one patient from each group harbouring M184I in DNA had the M184V mutation detected in RNA, whereas another from group 2 had the mutation E138K in both DNA and RNA. Defective viruses with in-frame stop codons, which are considered unable to replicate, were detected in DNA sequences in nine patients from group 1 and five from group 2.

Distribution of ARV resistance

Prevalence of resistance to individual ARVs is summarized in Figure 1(a) for group 1 and Figure 1(b) for group 2. When examining ARV classes, resistance to at least one of the following was observed in group 1 from RNA and DNA sequences, respectively: NRTI, 38 (51.4%) and 30 (40.5%); NNRTI, 24 (32.4%) and 16 (21.6%); and PI, 23 (31.1%) and 15 (20.3%). In the 19 patients with ≥2 RNA genotypic results prior to DNA genotype, a slightly lower proportion of drug resistance was observed using results from DNA compared with all previous RNA, respectively: NRTI, 8 (42.1%) and 15 (79.0%); NNRTI, 7 (36.8%) and 13 (68.4%); and PI, 8 (42.1%) and 10 (52.6%). In group 2, resistance to at least one of the following was observed from, respectively, RNA and DNA sequences: NRTI, 9 (34.6%) and 10 (38.5%); NNRTI, 9 (34.6%) and 10 (38.5%); and PI, 5 (19.2%) and 6 (23.1%).

Similarity in resistance information between DNA and RNA genotyping

In Table 2, the concordance in the number of agents for which resistance was observed, within each class of antiretroviral, was compared between DNA and RNA genotypes. Percentage of agreement was similar between groups 1 and 2 for NNRTI and PI classes, whereas group 1 had a roughly 20% lower percentage agreement than group 1 for NRTIs. Nevertheless, the ICC was much higher in group 2 compared with group 1 across all classes, with the smallest difference observed for PIs. There were no remarkable differences in percentage of agreement and ICC when patients harbouring defective viruses in DNA genotype were excluded from analysis (data not shown).

Net change in resistance information between DNA and RNA genotypes

The majority of patients had zero net difference in overall number of ARVs when resistance was observed: 46 (62.2%) in group 1 and 24 (88.9%) in group 2.

In group 1, more patients appeared to have more loss than gain in resistance information with respect to NRTI (31.1% versus 10.6%, Figure 2a left panel), NNRTI (17.6% versus 10.6%, Figure 2b left panel) and PI classes (18.9% versus 2.7%, Figure 2c left panel). In group 2, few patients exhibited loss of information in NRTI (3.7%, Figure 2a right panel) and NNRTI (7.4%, Figure 2b right panel) classes and no net change in resistance information was observed for PIs (Figure 2c right panel).

In patients from group 1, the net difference in resistance sum score was lower for those with longer duration of HIV infection (P = 0.006), longer ART duration (P = 0.002) and higher number of ARVs (P = 0.01). Only longer ART duration was significantly associated with a net difference in resistance sum score during multivariable analysis (Table 3). Loss of information was associated with older age (P = 0.01), longer time since nadir CD4+ T cell count (P = 0.004), longer known duration of HIV infection (P = 0.002), longer ART duration (P < 0.001) and higher number of ARVs (P = 0.003). Longer ART duration was the only determinant significantly associated with loss of information in multivariable analysis (Table 3). These results did not substantially differ when examining only patients harbouring drug-resistant strains (Table S2).

In group 2, any risk factor analysis on the differences in resistance information between RNA and DNA genotypes was precluded by the lack of distribution in the resistance sum score.

Discussion

In a population of HIV-1-infected patients of which roughly two-thirds had viral strains with drug resistance to at least one ARV, there was some degree of information loss on antiretroviral resistance in current DNA genotype compared with retrospective RNA genotypes when VL was low or undetectable. However, for some patients, DNA genotype does appear to give similar numbers of agents to which the strain is resistant compared with RNA. Higher concordance in DNA resistance information was also observed with RNA sequences from the same day versus several years in the past.

In the absence of RNA genotype due to low levels of replication, resistance testing of DNA is sometimes the only means by which clinicians are able to select appropriate ART.¹² DNA genotype has been retrospectively compared with RNA genotype in other studies, demonstrating somewhat good concordance between past RNA and DNA genotype in patients without viral failure⁹ and adequate comparability with respect to the number of DRMs between past RNA and non-defective DNA genotype.¹³ We build on these studies by demonstrating only marginally good comparability in the degree of antiretroviral resistance and whether information on resistance is lost with DNA genotyping, giving clearer guidance on the utility of this approach in identifying drugs to which resistance has developed.

In our analysis, we considered a statistic by which the extent of information loss from DNA to RNA genotype could be assessed. This approach is somewhat analogous to comparing genotypic susceptibility scores between sample sources, as employed in other studies.^14–16 A score of zero is the optimal goal and indicates that the DNA genotype provides the same type of information without considering specific ARVs. A negative score would indicate loss of information in the DNA sample compared with RNA, with lower values representing a greater degree of loss of information.

Our univariable analysis identified a number of risk factors associated with loss of information in the DNA genotype, which included longer time since nadir CD4+ T cell count, known duration of HIV infection, known duration of ART and number of cumulative ARVs since ART initiation. As these determinants are closely intertwined with ART duration, it is unsurprising that longer ART duration was the only factor associated with loss of information in multivariable analysis. ART duration could be a simple proxy for extensive HIV infection and potentially more complicated therapeutic history, leading to exposure to many ARVs and selective pressures to develop DRMs.¹⁷ Agents will be switched as treatment progresses and the selective pressures of acquiring certain mutations might no longer be present. In turn, previous DRMs could be naturally purged from the reservoir during longer periods of treatment.¹⁸ For instance, M184I/V in DNA is known to disappear in the absence of selection pressure from NRTIs, which could explain the effectiveness of lamivudine-containing dual therapies in patients with a history of M184I/V mutation.¹⁹

Nevertheless, there are other explanations for the information loss in DNA genotype observed in our study. Our study population with longer HIV infection and ART exposure could have had a higher frequency of virological failure, which has also been associated with poor concordance between RNA and DNA genotypes in other studies.^6,7 Higher person-years of viral replication, representing in part less effective ART combinations when levels are higher, did not correlate with loss of information. However, it is difficult to differentiate true virological failure from that caused by poor treatment adherence as an explanation for replicating HIV. Furthermore, the detection threshold from Sanger sequencing only allows detection of mutant variants as majority quasispecies and could miss DRMs representing less than 15%–25% of the viral population.²⁰ It could be that DRMs found in previous RNA samples could have been archived as minority variants, creating further loss of information. Ultra-deep sequencing detecting low-frequency variants harbouring DRMs has been found to aid prediction of antiviral resistance, but was not performed in our study.^21,22

Mutation profiles in archived DNA are also known to evolve despite long periods of sustained virological control^23,24 and hence could affect interpretation and comparability of sequences determined on DNA versus RNA genotypes. As the loss of resistance in DNA was associated with longer duration of HIV infection and longer history of being on ART, our results indirectly indicate that increased diversity of proviral DNA could be the reason for loss of resistance information. Clearance of archived DRMs in proviral DNA after long periods of virological control, lacking therapeutic selective pressures, have been recently reported with RT, which could imply residual replication of more fit viruses within some reservoirs.²⁵

For some patients, resistance to ARVs was able to be detected in the DNA genotype but not in the RNA genotype. APOBEC3 cytidine deaminases are known to induce G to A hypermutations in GG or GA dinucleotides in DNA, which could explain some of these cases.²⁶ Seven patients harboured APOBEC mutations E138K, M184I and M230I in HIV DNA without DRMs in RNA, consistent with other studies.^8,19 The potential significance of these mutations is still unknown; nevertheless, others have shown that recombination could occur between hypermutated sequences and circulating viral populations, eventually leading to the emergence of resistant viral strains.²⁷ Others have also found that concordance is affected in DRMs between DNA and RNA genotypes in case of defective proviruses.¹³ We were unable to corroborate these results, yet only a small proportion of patients in our study harboured stop codon mutations in DNA sequences and the study population was different.

For other cases in group 1, DRMs might have developed owing to a switch to another treatment after the most recent RNA genotype and thus the resistance observed in DNA might have captured selective pressures incongruent with those from past RNA genotypes.²⁸

Importantly, the ICC comparing resistance information from DNA to RNA genotype was much lower for NRTIs than NNRTIs or PIs, as demonstrated in another study evaluating detected resistance.²⁸ These results imply that any susceptibility to losing DRMs might occur more frequently with NRTI agents. NRTI is historically an older class of ARVs and was likely given during early infection among those with longer ART duration. The risk of observing archived NRTI resistance mutations that disappear over time could then be considered higher for these individuals.

Certain limitations of our study need to be addressed. First, the large heterogeneity of the studied population does make it difficult to generalize to other populations, such as those with less exposure to ART or HIV infection duration, or to more specific treatment regimens. This heterogeneity could have also increased variation, which likely affected some of our analyses. Second, the low sample size could have limited the power to detect certain risk factors associated with differences in genotype information between RNA and DNA. Third, genotypic results were obtained retrospectively and hence selection of patients was based on sample availability. Fourth, unlike many genotypic resistance scores, we considered intermediate resistance test results as resistance, which could be considered extreme for certain ARVs. Finally, resistance testing of INSTIs was not available at the time of study and was not included in our analysis. Other studies would be needed to confirm similar observations with this treatment class.

In conclusion, genotyping DNA has some degree of concordance compared with previous RNA genotypes, specifically when determining the extent of drug resistance, in a real-life setting of patients with low or undetectable VL during ART. It is already recommended that DNA genotyping could be carried out for patients who have no RNA genotype or for those patients without any knowledge of their ARV history.^4,15 Our results would contend that if neither ART history nor RNA genotype is available, DNA genotype could provide some useful information on drug resistance to the major classes of antiretrovirals. Nevertheless, patients having extensive years of treatment with multiple lines of ARV regimens might have more loss of information in their viral DNA genotype, making it more difficult to rule out ARV resistance.

Funding

This study was carried out as part of our routine work. A post-doctoral fellowship from SIDACTION was awarded to A. B. for part of the submitted work providing paid salary under a temporary contract with the Institut National de la Sante et de la Recherche Medicale (Inserm).

Transparency declarations

None to declare.

Supplementary data

Figure S1 and Tables S1 and S2 appear as Supplementary data at JAC Online.

References

Author notes